Mistral Small 2603
Bare model
The model alone, run by the benchmark harness with the fixed prompt.
- Vendor
- Mistral
- Provider
- Mistral La Plateforme
- Region
- EU
- Version
- mistral-small-2603
- Composition
- Model with the fixed benchmark prompt
F2 score
73.5%95% CI 67.7%–79.4%Leak recall
73.6%95% CI 66.5%–80.8%Over-redaction
26.7%95% CI 22.1%–33.2%Output precision
58.3%95% CI 50.6%–63.6%Cost per 1,000 pages
EUR 0.5095% CI not applicable · measured list-price costRecall by bucket
Universal identifiers81.5%95% CI 74.3%–92.2%
Names96.0%95% CI 94.5%–97.3%
Contextual personal data46.2%95% CI 32.6%–57.8%
Breakdowns
Groups below the reporting floor are not shown.
| Language | Documents | F2 | Recall | Over-redaction |
|---|---|---|---|---|
| de | 24 | 71.8% | 78.5% | 46.5% |
| en | 24 | 70.2% | 72.8% | 38.6% |
| es | 22 | 76.1% | 76.1% | 23.7% |
| fr | 20 | 73.9% | 74.5% | 28.2% |
| it | 23 | 64.4% | 61.9% | 23.2% |
| nl | 22 | 79.1% | 80.3% | 25.3% |
Provenance
- Dataset version and split
- 1.0.0-r3+labels.20261006.1 · tuning
- Scoring
- 1.3
- Harness version
- pilot-2026-10-15
- Run date
- 2026-10-14
- Documents
- 135
- Pages
- 892
- Repeats
- 1
- Price date
- 2026-09-29
- Spend
- EUR 0.44
- Credits
- None · paid by OpenDisclosure
Cite this
@misc{eu-redaction-mistral-small-2603-mistral-eu-no-roster,
author = {OpenDisclosure},
title = {EU Redaction Benchmark: Mistral Small 2603},
year = {2026},
url = {https://opendisclosure.eu/benchmark/models/mistral/mistral-small-2603/no-roster/},
note = {r3-tuning-pilot-2026-10-15; tuning split; scoring 1.3}
}Download JSON